U.S. patent number 3,853,496 [Application Number 05/319,991] was granted by the patent office on 1974-12-10 for method of making a metal insulator silicon field effect transistor (mis-fet) memory device and the product.
This patent grant is currently assigned to General Electric Company. Invention is credited to Manjin J. Kim.
United States Patent |
3,853,496 |
Kim |
December 10, 1974 |
METHOD OF MAKING A METAL INSULATOR SILICON FIELD EFFECT TRANSISTOR
(MIS-FET) MEMORY DEVICE AND THE PRODUCT
Abstract
The present invention relates to the method of making a memory
device of the metal insulator silicon field effect transistor
structure having a gate region combining a chemically formed thin
oxide layer and a second insulating layer, such as silicon nitride,
and to the novel product which results from this method. The method
entails the step of chemically oxidizing the surface of the silicon
channel region by a self-limiting process to form a thin porous
oxide, and a nitriding step which is conducted under conditions
producing optimum interface traps and minimum initial charge. Both
steps lead to highly reproducible devices. The method is readily
applied to large arrays of devices, offering ease of manufacture
and close device parameter control. One form of the process
provides for the production of both memory and read-out devices
appropriate in an array.
Inventors: |
Kim; Manjin J. (Liverpool,
NY) |
Assignee: |
General Electric Company
(Syracuse, NY)
|
Family
ID: |
23244402 |
Appl.
No.: |
05/319,991 |
Filed: |
January 2, 1973 |
Current U.S.
Class: |
438/287;
257/E21.209; 257/406; 257/E21.288; 257/E21.283; 257/E29.309 |
Current CPC
Class: |
H01L
21/02271 (20130101); H01L 21/02211 (20130101); H01L
29/792 (20130101); G11C 16/0466 (20130101); H01L
21/31675 (20130101); H01L 21/022 (20130101); H01L
21/02255 (20130101); H01L 21/31654 (20130101); H01L
21/02238 (20130101); H01L 21/0217 (20130101); H01L
29/40114 (20190801) |
Current International
Class: |
H01L
21/02 (20060101); G11C 16/04 (20060101); H01L
21/316 (20060101); H01L 21/28 (20060101); H01L
29/792 (20060101); H01L 29/66 (20060101); H01l
011/14 () |
Field of
Search: |
;29/571,578 ;117/118
;11/201 ;317/235 |
References Cited
[Referenced By]
U.S. Patent Documents
Primary Examiner: Lake; Roy
Assistant Examiner: Crouse; R. Daniel
Attorney, Agent or Firm: Lang; Richard V. Baker; Carl W.
Neuhauser; Frank L.
Claims
I claim:
1. The method of making a metal insulation silicon field effect
transistor (MIS-FET) memory device, comprising the steps of:
a. forming an insulating SiO.sub.2 layer over a low resistivity
substrate,
b. etching openings in said SiO.sub.2 layer and diffusing in source
and drain regions,
c. sealing said source and drain regions with SiO.sub.2 and opening
an intermediate channel region,
d. immersing the substrate in concentrated nitric acid to form a
low temperature chemical oxide of silicon over said channel region
under self-limiting conditions in order to produce a first
insulating layer about 20A in thickness,
e. forming a second insulation layer over said chemical oxide to
provide a region of deep traps at their interface,
f. opening said source and drain contact regions, and
g. metallizing said source and drain regions and said second
insulating layer over said channel region.
2. The method of claim 1 wherein during formation of said chemical
oxide, the concentration of said nitric acid is in excess of 60
percent by weight and maintained at a constant temperature, said
reaction continuing for at least 20 minutes.
3. The method of claim 2 wherein said constant temperature is set
at the boiling point of said nitric acid.
4. The method of claim 3 wherein said acid is 84 percent by weight
and said boiling point is 86.degree.C.
5. The method of claim 1 wherein said second insulating layer is
silicon nitride, said formation occuring by reacting a mixture of
NH.sub.3 with a small amount of SiH.sub.4 at a temperature of from
750.degree.C to 850.degree.C to minimize the fixed charge therein,
and the upper limit favoring the formation of the amorphous form of
nitride, and avoiding crystallinity
6. The method of claim 1 wherein said second insulating layer is
Al.sub.2 O.sub.3 formed by hydrolyzing AlCl.sub.3 in a mixture of
H.sub.2 and CO.sub.2 at 900.degree.C.
Description
BACKGROUND OF THE INVENTION
1. Field of the Invention
The present invention relates to a method of making MIS-FET memory
devices which are capable of storing a charge in a double layer
insulated gate region. The method more particularly relates to the
method of forming the two insulating layers, and of improving the
method for manufacture of the memory device to accommodate
fabrication of readout elements as well. The invention also relates
to the product of the process, MIS-FET memory devices.
2. Description of the Prior Art
The inventive product may be regarded as structurally and
functionally related to the conventional MOS-FET, of which the
enhancement mode P-channel MOS-FET is an example. Such devices are
described in the Motorola Data Book, Fourth Edition, pages
AN47-AN57. Such a device comprises an N-silicon substrate of high
resistivity (5 ohm cm, typically) with source and drain regions of
high conductivity (50 ohm per square, typically) P-silicon. A gate
region is formed in the substrate between the source and drain. The
gate comprises an insulating layer of SiO.sub.2 applied over the
gate region, and a thin layer of Si.sub.3 N.sub.4 over the
SiO.sub.2 for passivation. Electrode metallizations are applied to
the source, drain and gate.
The FET transistor resembles a vacuum tube with a control grid in
that the gate control of the drain current is by an electric field,
and there is a threshold at which conduction begins. With customary
potential differences between source and drain, and "zero" gate to
substrate potential, the device appears to be a pair of back to
back diodes, and no current flows. With a negative potential on the
gate, field induced positive holes appear at the upper surface of
the channel until the region beneath the oxide becomes effectively
a p-type semiconductor region, permitting current to flow between
source and drain through the induced "P channel". The threshold of
conduction occurs at the point where the concentration of minority
carriers at the surface equals the impurity concentration in the
bulk semiconductor. The current flow between source and drain
increases with increased negative gate potential, leading this mode
of operation to be characterized as the "enhancement mode". The
oxide layer of conventional MOS-FETs is normally thick enough to
remain insulating under the customary ranges of applied gate
potentials and the nitride is normally very thin, serving primarily
as a passivation layer to protect the under layers from sodium
contamination.
The insulated gate field effect transistor was originally used
primarily for linear signal operation. However, when the gate oxide
was replaced or combined with other insulators, as for instance
nitride (Si.sub.3 N.sub.4), it was found that the MIS (Metal
Insulation Silicon) - FET devices as they are generically called,
often exhibited a new property. In the case of the metal nitride
oxide field effect transistor (MNOS-FET), the conduction
characteristic is a slightly curving function of the gate
potential, curving more in the region of the threshold and
approaching linearity at higher values. For small ranges of
variation in gate potentials, the conduction characteristic is
retraceable. For larger values of gate voltage, the conduction
characteristic is shifted with high gate voltages. This
displacement is explained as arising from charges drawn into the
insulating layers, where they become trapped and influence
conduction in the channel region. The effect displaces both the
"linear" portion and the threshold region of the conduction
characteristic. The displacement can be removed by either applying
a strong opposite (positive) gate potential to remove the trapped
charges or by waiting a long enough time for them to leak out
naturally. In conventional linear MIS-FETS, the hysteresis property
is small, and unoptimized.
The variable threshold effect led to the development of MIS-FET
devices optimized as memory devices. Since the threshold of
conduction is distinct in FETs, the unoptimized devices could be
used in non-linear or switching modes for active logic
applications. When it became apparent that the position of their
thresholds could be varied by control of charges trapped in the
gate region, interest grew in using this phenomenon for passive
logic applications, i.e. data storage. Since the trapped charges
did not require external fields for their continuance, they
promised an electronic memory not requiring the continuous supply
of energy for memory retention. The insulation would, however, have
to be good enough to keep the "trapped" charges long enough. The
MNOS-FET and the MAOS-FET devices are two variable threshold
devices resulting from this activity. In the MAOS-FET device the
gate region also uses two distinct insulating layers, a silicon
oxide layer and an aluminum oxide (Al.sub.2 O.sub.3) layer and
trapping is accomplished at their isolated interface.
MIS-FET memory devices are widely reported, but continue to exhibit
great variability. Frohman-Bentchkowsky and Lenzlinger, in an
article entitled "Charge Transport and Storage in
Metal-Nitride-Silicon MNOS Structures", Journal of Applied Physics,
Volume 40, pp. 3307-3319, 1969, proposed an oxide layer of 50-200A
thickness and a nitride layer of 200-1000A thickness. A thinner
oxide 15-35A was proposed by J. T. Wallmark and J. H. Scott, in an
article entitled "Switching and Storage Characteristics of MIS
Memory Transistors", in the RCA Review, Volume 30, pp. 335-365,
1969. A similar device was discussed by E. C. Ross and J. T.
Wallmark in an article entitled "Theory of the Switching Behavior
of MIS Memory Transistors" in the RCA Review, Volume 30, pp.
336-381, 1969.
With varying degrees of successes reported, and some convergence in
view toward a common understanding of the underlying mechanisms,
MNOS-FET, MAOS-FET storage devices do not yet appear to be
commercially successful. One objection appears to have been the
difficulty in achieving reproducible characteristics. Since storage
devices are most useful when assembled into large arrays,
non-reproducibility has been a major objection to their use.
Furthermore, although a better understanding of the internal
physical mechanisms has led to improved device parameters, reported
devices have not yet been optimized. Some have overly large
thresholds; others have tended to show a deterioration in memory
with large amounts of memory cycling, others have suffered from
objectionably large decays (with time) in their threshold
separations. Thus, improvement in the processing has been needed
both to optimize the parameters of the finished devices and to
reduce their variability. Furthermore, since the memory elements
have been desired for use in arrays, processing had been needed
that would lead to ease in fabrication of both the storage and
necessary decoding elements.
SUMMARY OF THE INVENTION
Accordingly, it is an object of the present invention to provide an
improved method of making an MIS-FET memory device.
It is a further object of the present invention to provide an
improved method of making an MIS-FET memory device having high
reproducibility.
It is another object of the present invention to provide an
improved method of making the double layer insulated gate region of
a MIS-FET device.
It is a further object of the invention to provide an improved
method of making a MIS-FET memory and decoding device suitable for
use in an array.
It is an additional object of the invention to provide an improved
MIS-FET device having an improved gate region.
It is another object of the invention to provide an improved
MIS-FET device having a structure particularly suited for ease in
manufacture.
These and other objects of the invention are achieved in a method
of making a metal insulation silicon field effect transistor
(MIS-FET) memory device, comprising the steps of forming an
insulating SiO.sub.2 layer over a low resistivity substrate;
etching openings in the SiO.sub.2 and diffusing in source and drain
regions; resealing the source and drain regions with SiO.sub.2 and
opening the channel region; immersing the substrate in concentrated
nitric acid to form a low temperature chemical oxide over said
channel region under self-limiting conditions to a thickness of
about 20A; forming a silicon nitride layer of from 300 to 1000A
over the chemical oxide by reacting a mixture of NH.sub.3 with a
small amount of SiH.sub.4 (1/7500) at a temperature of from
750.degree. to 850.degree.C to minimize the fixed charge therein;
opening the source and drain contact regions; and metallizing the
source and drain regions and said silicon nitride layer over said
channel region.
More particularly, the nitric acid used in the chemical oxide
reaction is concentrated, lying in the range of from 60 - 84% by
weight, and conveniently held at the boiling point of the acid
(86.degree.C in the case of 84% concentration) during the reaction.
The reaction is carried into self-limiting, the period normally
required being some time in excess of 20 minutes. The nitride layer
is formed using an inert carrier gas including nitrogen, and the
range specified includes 830.degree.C, the point at which fixed
charge in the nitride layer becomes zero under these conditions.
The upper limit, 850.degree.C is selected to avoid crystallinity,
and to favor the amorphous form of nitride. The chemical oxide
prepared in this manner is porous, and has a large number of
physical defects forming both deep and shallow traps. The nitride
formed over the oxide leads to the formation of a large number of
deep traps at their interface, the number arrived at under the
indicated circumstances being about 6 - 8 .times. 10.sup.12
traps/cms.sup.2. The fixed charge under these circumstances is less
than 1 .times. 10.sup.12 per /cm.sup.2.
Alternatively, an Al.sub.2 O.sub.3 layer may be substituted for the
nitride layer. When this is done the nitride step is replaced by
hydrolyzing AlCl.sub.3 in a mixture of H.sub.2 and CO.sub.2 at
900.degree.C. The Al.sub.2 O.sub.3 layer so formed is then allowed
to build up to a relatively large thickness in relation to the
oxide layer (>300 A) as in the nitride process.
The process may be extended to permit the formation of a stable
gate device to accompany the variable gate device. Both kinds of
devices are normally required in the formation of arrays. When this
is done, a conventional "thermal" oxide step is added just before
the chemical oxide step. The thermal oxide is prepared at a higher
temperature of 1000.degree.C and results in an oxide layer
(SiO.sub.2) typically of 500A thickness.
The MIS-FET which results from this process comprises a low
conductivity silicon substrate of one conductivity type and forming
a first connection region; a pair of high conductivity diffusions
of the other conductivity type spaced to form an intervening narrow
width channel region; source and drain metallizations applied to
the diffused regions to form corresponding source and drain
electrodes; and a gate comprising a chemical oxide of silicon of
approximately 20A thickness grown on the surface of the substrate
over the channel region for providing a tunneling layer, a second
insulating layer whose conduction is substantially less than in
said chemical oxide layer and providing a region of deep positive
traps at their interface, and a gate metallization applied to the
second insulating layer over said channel region to form a gate
electrode.
The chemical oxide layer has the properties outlined above in
connection with the process description and is highly reproducible.
The second insulating layer may take the form of a nitride
(Si.sub.3 N.sub.4) layer or an aluminum oxide (Al.sub.2 O.sub.3)
layer. The finished device is designed to function with the thermal
oxide forming the conduction region under high field conditions,
while the second insulation layer has minimal conduction, minimum
fixed charge, and at its interface with the chemical oxide provides
a maximum number of deep traps for establishing the desired
variable threshold action. Normally, conduction in the oxide layer
is by tunneling. The mode of operation can either be enhancement or
depletion mode, and the devices may be either "P" channel or "N"
channel.
The preferred embodiment uses a nitride second insulating layer, in
a P channel device, with the oxide supporting a tunneling mode of
conduction under high field conditions.
BRIEF DESCRIPTION OF THE DRAWING
The novel and distinctive features of the invention are set forth
in the claims appended to the present application. The invention
itself, however, together with the further objects and advantages
thereof may be best understood by reference to the following
description and accompanying drawings in which:
FIG. 1 is an illustration of a MIS-FET memory and decode device in
accordance with the invention; and
FIG. 2(a) through 2(h) are illustrations of Applicant's novel
method of making a MIS-FET memory and decode device, each
characterizing a different stage in the fabrication.
DESCRIPTION OF THE PREFERRED EMBODIMENT:
Referring now to FIG. 1, a finished metal insulation silicon field
effect transistor (MIS-FET) memory and readout device is
illustrated. The insulation layers specifically illustrated are
nitride and oxide. The finished structure is one that is readily
reproduced in quantities and one which may form an element in a
large array. The illustrated pair of FET devices 11 and 12 are
P-channel devices formed on a substrate 13 of silicon of low n
resistivity, normally in the range of from 1 to 20 ohm centimeters.
In the memory element 11, a first P+ diffused region at 14 is the
source diffusion and spaced to the right at 15 is a second P+
diffused region for the drain. In the readout element 12, a first
P+ diffused region 16 is the source diffusion, and spaced to the
right is a second P+ diffused region 17 for the drain. All four
regions are diffused with Boron to about 50 ohms/sq. Source and
drain diffusions are spaced about 0.2 mils apart and thus define
two channel regions in the substrate.
The surface of the silicon substrate is covered with two insulating
layers which are opened over the diffused and channel regions. The
lowermost layer 22 is a relatively thick layer (1.1 microns) of
SiO.sub.2 having good insulating qualities, and it is covered by an
appreciably thinner layer 23 (300 to 1000A) of Si.sub.3 N.sub.4,
also having good insulating qualities. The nitride layer, as will
appear, serves both to passivate the underlying silicon, and in the
channel region, is a part of the gate insulation. The electroding
18, 19, 20, 21 to the source and drain regions contact the
substrate through the openings in the insulating layers. The
contacts each comprise a thin initial deposition of titanium (200A)
followed by a more substantial deposition of aluminum (7000A). The
contacts are alloyed to the underlying diffused region. As so far
described, the source and drain constructions, and the doping
levels are conventional.
The insulating gates, while unique in several respects, are of a
conventional configuration. The memory gate comprises a double
layer whose undermost layer is a unique, chemically formed thin
(20A) oxide 24. The upper layer is the Si.sub.3 N.sub.4 layer 23
previously noted. The gate is electroded by a Ti-Al metallization
like that for the source and drain. The gate metallization is
applied on top of the two layers as shown at 25. In the readout
device 12, the stable gate comprises a thicker (500A) underlayer 26
of thermal oxide, the Si.sub.3 N.sub.4 layer, and finally a similar
Ti-Al metallization 27 forming the gate electrode.
The memory gate oxide is normally about 20A units in thickness,
although it can be made slightly thinner. While the thickness
selected is not unusual, the nature of the oxide is. It is
chemically unusual in that the oxide is approximately SiO, the
oxygen being present in quantities slightly larger than 1 to 1. The
oxide, when formed by reaction with concentrated nitric acid at
86.degree.C, and in the manner to be described more fully below,
appears to have many voids of a near molecular size and to have a
generally porous and irregular structure. The indicated reaction
process permits easy control of thicknesses at 20A and slightly
below. An ellipsometer reading indicates the index of refraction of
the chemical oxide layer to be about 2.4 as opposed to 1.6 for
conventional SiO.sub.2.
The Si.sub.3 N.sub.4 layer, as will be described more fully below,
is also applied in conventional thicknesses, normally from 300 to
1000A. The deposition involves the pyrolytic reaction between
NH.sub.3 and SiH.sub.4 in trace quantities (1 part in 7500), with a
nitrogen carrier and no hydrogen present. The temperature is
carefully controlled to a value between 750.degree.C and
850.degree.C. This temperature produces a minimum initial charge
density -- and accordingly a minimum gate threshold -- without
losing other desirable nitride properties. At the same time, the
temperature selection allows the nitride to form on the chemical
oxide without substantial deterioration of the oxide and with the
formation of a near optimum number of deep traps.
A device which is made in the manner specified, using the indicated
thin chemical oxide 20A in combination with the Si.sub.3 N.sub.4,
exhibits no measurable loss in storage capability from repeated
cycling (10.sup.9), a low threshold (-3 volts in a typical case),
and a threshold separation of -7 volts. The value is held for a
year at 20% drop.
The immunity to degradation from cycling of the present device is
believed to lie in the nature of the chemical oxide layer. Some
correspondingly thin thermal oxide layers have been reported to
have shown degradation after 10.sup.5 cycles. The chemical oxide
has not shown observable degration after 10.sup.9 cycles. The
chemical oxide is formed at a low temperature (86.degree.C) in a
manner which permits a larger number of physical defects than in
high temperature oxidation. These defects consist of both deep and
shallow density, such as voids, lattice vacancies, vacancy
clusters, dangling bonds, etc. Many of these defects are of a kind
that continued tunneling would tend to induce. Thus, in the
chemical oxide, where there is near saturation in defects of this
kind, the tunneling which accompanies cycling of the memory cell is
not believed to produce significantly more. This is believed to
explain why repeated cycling produces no observable changes in the
operation of the chemical oxide layer.
By contrast, the thermal oxide layer appears to alter slightly with
the repeated tunneling accompanying cycling. These produce
additional shallow defects, which, while not harmful in an initial
device, perceptibly alter its electrical characteristics in the
memory unit being cycled. Thus, in an array, if some units are
altered while others are not, their non-uniformity creates a severe
operational problem, jeopardizing the entire array.
As previously noted, the oxide layer that is used in the memory
element is formed by reacting the silicon substrate with
concentrated nitric acid (at 86.degree.C), the boiling point of the
acid, for about 20 minutes. This method of treatment leads to the
formation of a thin oxide, and one that is of reproducible
thickness and electrical properties.
The method involves treatment of the silicon surface with a
concentrated nitric acid solution immediately after cleaning. The
process produces a repeatable layer by careful selection of the
processing conditions; temperature, acid purity and concentration,
and time. The very low temperature and long time (20-30 minutes)
involved facilitates control of the process.
The HNO.sub.3 oxidation process is believed to proceed via an
intermediate series of autocatalytic reactions in which HNO.sub.2
is first formed by reactions between N.sub.2 O.sub.4, NO and water
as follows:
N.sub.2 O.sub.4 + H.sub.2 O .fwdarw. HNO.sub.2 + HNO.sub.3 (1)
N.sub.2 O.sub.4 + NO + 2 H.sub.2 O .fwdarw. 4 (2) .sub.2
it is also well known that surface atoms of silicon are active so
that at room temperature in air they form oxides as large as 8A.
The oxidation step in the film growth process involves the
formation of positive Si ions:
Si .fwdarw. Si.sup.+.sup.+ + 2e.sup.- (3)
In concentrated HNO.sub.3, the potential difference between the
silicon and negative electrolyte is as much as 0.5V and the
ionization and ion migration through the thin oxide will be
enhanced by this potential. The electrical field of the silicon
interface falls very rapidly, and is believed to operate first to
attract very strongly, and then less strongly oxidizing ions into
contact with the less mobile Si ions. The silicon ions thus react
with HNO.sub.2, which is a more active oxidizing agent than
HNO.sub.3, according to the reaction:
2 HNO.sub.2 .fwdarw. H.sub.2 O + NO + 20.sup.-.sup.- (4)
Si.sup.+.sup. + + O.sup.-.sup.- .increment. SiO (5)
this produces a monoxide rather than the dioxide, although the
compositon is not believed to be stoichiometric, but to vary
slightly about 1 to 1.
Like chemical stain films on silicon, the rate of film growth, for
the initial state of oxidation, appears to follow a quadratic
form:
dx/dt .varies. k.sub.R /2x (6)
where x is the thickness, and K.sub.R the rate constant, depending
on the temperature:
k.sub.R .varies. exp(-E.sub.a /kT) (7)
where E.sub.a is the thermal activation energy. For low
concentrations, the initial oxidation rate is also proportional to
the molar concentration of HNO.sub.3 :
dx/dt .varies. C.sup.n exp(-E.sub.9 /kT) (8)
where C is the concentration of HNO.sub.3 in moles per liter, n is
the reaction order.
Typically, a treatment time of 20-30 minutes in concentrated
HNO.sub.3 was selected. To stabilize temperature, the acid was
boiling (86.degree.C) and refluxed to insure constant
concentration. Electronic grade HNO.sub.3, specifically free of
sodium was used rather than reagent grade acid.
The reaction is readily controlled because the conditions noted
above lead to self-limiting. Applicant's experiments show that with
concentrated nitric acid, the reaction proceeds at the parabolic
rate for about 10 minutes. After this period the oxide build up
begins to slow down, and appears to come to a practical stop after
about 20 minutes from start. The final thickness is slightly
dependent upon the concentration of the acid. Operation with lesser
concentrations of nitric acid than 60% have led to difficulty in
achieving reproducibility. Concentrations from 60 percent to 84%
lead to reproducible thicknesses of the oxide layer. The nitric
acid reaction permits the formation of layers that are
substantially thicker than the residual oxide. An oxide thickness
of 20A versus 15A represents a difference in storage time between
years and weeks. Usage below 15A is thus less promising from both
the matter of reproducibility and non-volatility.
Use of the boiling temperature (86.degree.C) is not critical,
except that temperature also influences the initial process rate.
Thus, the process may be conducted at the particular boiling point
of the concentrations of acid used or at some regulated
temperature. At a lower temperature an additional allowance of time
for the reaction to self-limit may be necessary (30 minutes versus
20 minutes).
As earlier noted, the oxide layer is thin, approximately 20A, while
the nitride layer is widely variable and thicker (300 - 1000A).
This oxide thickness is thin enough to be fast and is readily
reproduced with substantial uniformity. If a slower or faster, or
higher or lower voltage gate action is sought, most control can be
obtained by adjusting the thickness of the nitride layer.
Standardizing the oxide layer has tended to standardize the
conduction and discharge phenomena since the nitride is made thick
enough to force conduction through the oxide at all times. At the
interface of the two layers, differences in lattice dimenison,
irregularities and assorted physical property mismatches form the
large number of deep traps responsible for the storage phenomena.
This remains the same irrespective of the thickness of the nitride
layer.
In the oxide layer, avalanche injection occurs from the substrate
under high field polarity conditions. Assuming an n-p condition,
electrons can be injected into the oxide conduction band. Tunneling
occurs throughout the forbidden energy gap when the oxide is thin
as herein contemplated. The dominant conduction mechanism is
Fowler-Nordheim emission, itself dependent on the presence of
thermally excited electrons to exceed the barrier.
The nitride, on the other hand, is sufficiently thick (even at
300A) so that tunneling is forbidden and Poole-Frankel conduction
is predominant, i.e. field enhanced thermal excitation of trapped
electrons into the nitride conduction band. Calculations show that
the oxide conduction is 5.4 .times. 10.sup.-.sup.4 amperes/cm.sup.2
at a field of 8 .times. 10.sup.6 V/cm (zero stored charge. Also
assuming a stored charge at the interface of 6.0 .times.
10.sup.+.sup.12 states/cm.sup.2, the oxide current is reduced to
1.46 .times. 10.sup.-.sup.13 amperes/cm.sup.2. Thus, as the
interface approaches saturation, the oxide charging ceases. On the
other hand, the nitride current using a field of b 3.7 .times.
10.sup.6 V/cm is 1.1 + 10.sup.-.sup.10 amperes/cm.sup.2. Thus,
during charging the nitride current is negligible, in relation to
current through the oxide. Analysis of the decay currents also
shows that the charge decay is through the oxide layer rather than
the nitride.
The predicted displacement in threshold voltages is
approximately:
V.sub.T = Q.sub.I (t)/C.sub.N Where Q.sub.I (t) is the time
dependent charge density in the oxide-nitride traps, and C.sub.N is
the nitride capacitance per square centimeter. Typical observed
separations are 7 to 8 volts, the stored charge corresponds to 7
.times. 10.sup.12 traps per cm.sup.2, and the threshold is below 3
volts.
The technique for achieving threshold voltages as low as one volt
is largely dependent upon the nitride processing. The threshold
voltage, V.sub.T, depends on the substrate carrier concentration,
the amount of fixed charge in the dielectric, and the difference in
work function between the silicon and metal (.phi..sub.MS)
involved. Since V.sub.T for P-channel devices is relatively
insensitive to concentration ranges below 10.sup.15 atoms/cm.sup.3,
and since .phi..sub.MS is fixed for a given metal system, V.sub.T
depends primarily on the initial fixed charge density (Q.sub.IC) in
the two dielectric layers. Since the oxide layer is thin in
relation to the nitride, and the initial fixed charge in the oxide
made as small as possible, the remaining and more important control
of the initial fixed charge in the gate insulation is in the
formation of the nitride layer.
To obtain a minimum initial charge, we found that a high NH.sub.3
/SiH.sub.4 ratio is needed and the carrier should be an inert gas
including nitrogen. Hydrogen should not be present. These steps
reduce positive silicon or hydrogen ions in the nitride. However,
the most important parameter controlling the value of the initial
charge is the deposition temperature. This effect has been reported
in the Journal of Applied Physics, 15 Dec. 1969, pages 408, 409, in
a letter by E. C. Ross et al.
The deposition temperature appears to cause a slight departure in
stoichiometry between the reacting elements. At the lowest
temperatures an excess of silicon ions occurs resulting in very
deep "+" states. These states are too deep for conduction, but act
to neutralize the gate field. As the deposition temperature
increases, this "initial" charge falls until it crosses through
zero at 830.degree.C and gradually assumes negative values. The
number of states remains low over a substantial temperature range
>1 .times. 10.sup.12 (+) atoms/cm.sup.2 at 750.degree.C and
>1 .times. 10.sup.12 (-) atoms/at 1000.degree.C. At temperatures
above 900.degree.C, however, the deposition creates problems in
deposition of crystalline Si.sub.3 N.sub.4 particles while the
amorphous form is desired, and the possibility of deterioration of
the underlying chemical oxide layer. The observed maximum charge
appears to be quite high throughout this region (from 6 - 8 .times.
10.sup.12 states/cm.sup.2). Thus, for optimization of the minimum
threshold, the region about 830.degree.
is preferred. The usable range appears to be from 750.degree.C to
about 850.degree. C.
To complete the process of making the MIS-FET memory and readout
element is illustrated in FIGS. 2(a) through 2(h). Five masks are
required in the process.
The starting material for the mis-fet device is typically a 2 ohm
centimeter, N-type silicon wafer with a [100] crystal orientation.
The material is normally within the range of 1 - 20 ohms cm. The
wafer surface is optically polished and carefully cleaned before
starting the initial oxidation. The cleaning starts with a solvent,
then aqua regia; ammonium hydroxide (NH.sub.4 OH) and hydrogen
peroxide (H.sub.2 O.sub.2); and finally 5% hydrofluoric acid. After
this, the surface is rinsed in distilled water having a purity of
in excess of 10.sup.6 ohm centimeters. After rinsing, the substrate
is ready for the initial oxidation.
The initial oxidation is a thermal oxidation which provides a layer
of SiO.sub.2 to a uniform depth of 1 micron over the top of the
wafer. The oxidation is carried on at 1100.degree.C, using steam
for 2 hours. At the end of the oxidation the substrate appears as
illustrated in FIG. 2(a).
The SiO.sub.2 layer is then etched through the SiO.sub.2, using a
first mask to form the source and drain openings as shown in FIG.
2(b). The mask is a photoresist mask applied and formed in place in
a conventional manner. The etching of the SiO.sub.2 takes place at
about room temperature (20.degree.C) using 15% solution of
hydrofluoric acid for 10 minutes.
After etching, the exposed silicon is carefully cleaned preparatory
to the boron diffusion of the source and drain regions. The
cleaning procedure normally involves some of the same steps as the
initial cleaning procedure; aqua regia, ammonium hydroxide, 5
percent hydrofluoric acid and rinsing in distilled water.
With clean openings as shown in FIG. 2(b), the borom diffusion is
undertaken to form source and drain diffusions. This step takes
place at a wafer temperature of 1120.degree.C for 1 hour in a
diffusion furnace. The boron source is a mixture of 0.25% BCl.sub.3
in N.sub.2 and controlled to achieve a sheet resistivity of 50
ohms/sq. Small quantities of H.sub.2 and O.sub.2 are also normally
present in the source. The diffusion is conventional and may use
other procedures.
The source and drain diffusions are then sealed with a 6000A layer
of SiO.sub.2. The oxidation takes place with an initial dry O.sub.2
at 1120.degree.C for 10 minutes, then in steam at 1120.degree.C for
30 minutes, followed by dry O.sub.2 at 1120.degree.C for 30 minutes
more. The duration and temperature of this step is designed to
drive in the boron diffusion and to narrow the channels between
sources and drains. At this point the substrate is as illustrated
in FIG. 2(c).
Using a second mask, and a conventional photoresist process, the
SiO.sub.2 is now opened up over the channels in the readout
element. The opening must expose narrow regions to either side of
the channel as illustrated in FIG. 2(d). The process entails
exposure to 15% HF at 20.degree.C for 10 minutes through a
photoresist mask. After the etching, the photoresist is removed and
the surface is cleaned by the cleaning procedure earlier
mentioned.
The exposed channel region for the readout element is then oxidized
from a thick thermal oxide. The process is conventional, using dry
O.sub.2 at 1000.degree.C and is typically timed to form a layer of
500A thickness.
After the formation of the "thick" oxide, the thin chemical oxide
layer is formed in the memory element. This requires a third mask
providing openings over the channel region, and a repetition of the
etching and cleaning steps.
The thin oxide gate is formed as shown in FIG. 2(e). The substrate
is submerged in boiling (86.degree.C) nitric acid (electronic
grade) for about 20 minutes. A thin layer of oxide of about 20A is
formed over the exposed silicon. The composition, as has been
explained, due to the low temperature formation, and disassociation
activity of nitric acid forms a near Si.sub.1 O.sub.1 composition,
rather than SiO.sub.2.
Nitride is then deposited over the substrate to a thickness of 300
to 1000A to form the nitride layer for both gate regions. According
to one procedure, it is formed by the pyrolysis of SiH.sub.4 with
NH.sub.4 at 850.degree.C using a carrier of argon or nitrogen.
H.sub.2 should not be present. The ratio of NH.sub.3 to SiH.sub.4
is 7,500 to 1. The nitride deposits at the rate of about 180A per
minute, and the process only need run for about 5 minutes. During
the deposition, which uses a vertical reactor heated by an r.f.
generator, the wafers are rotated to allow for uniform deposition
into the openings. The nitride must be particularly clean, and the
use of high purity gas sources essential. At this temperature, the
nitride layers remain amorphous and minimum threshold devices
result. At this stage, the device is as shown in FIG. 2(f).
The nitride formed over the thermal oxide in the "stable" gate and
over the chemical oxide in the variable threshold gate forms the
second insulating layer in both the readout and memory devices. In
the memory device, its presence creates a large trap density in the
SiO-Si.sub.3 N.sub.4 interface essential to variable threshold
operation.
The nitride layer is useful for passivating the SiO.sub.2
generally, but must now be removed over the source and drain
regions before the electroding. The nitride is inert to most
chemicals and its localized removal somewhat difficult. The
preferred removal method is to use a fourth photoresist mask with a
supplementary deposition of SiO.sub.2. With the photoresist mask in
place covering over the source and drain regions, SiO.sub.2 is
first sputtered to a depth of 1000A over the surface of the
substrate. The sputtered SiO.sub.2 has good adhesion to the
Si.sub.3 N.sub.4. Next, 3000A of SiO.sub.2 is deposited using a
conventional siloxane deposition at 400.degree.C. The double
SiO.sub.2 layer avoids undercutting and is pinhole free. The
photoresist and the SiO.sub.2 covering it are then removed from the
source and drain windows to expose the Si.sub.3 N.sub.4 layer for
its removal as shown in FIG. 2(g).
The Si.sub.3 N.sub.4 layer is etched away using H.sub.3 PO.sub.4
acid at 80.degree.C for 15 minutes. This exposes the thin 6000A
layer of SiO.sub.2 over the sources and drains. Using 15% HF for 6
minutes, the SiO.sub.2 mask (4000A) and the 6000A double layer
SiO.sub.2 over the source and drain regions are etched away. This
stage of the substrate is illustrated in FIG. 2(h). The etching
exposes the diffused source and drain regions on the substrate. The
HF etching stops at the nitride layers protectively covering the
earlier SiO.sub.2 layers. After cleaning, the exposed source and
drain regions are now ready for metallization.
Electrode metallization of the source, gate and drain regions is
then applied in a conventional manner using an electron beam
evaporator and the fifth mask. The preferred metallization involves
a thin (200A) initial layer of titanium, followed by an aluminum
deposition of 7000A carried on at a substrate temperature of 100C.
After removal of the mask, electrodes are "alloyed" in a nitrogen
atmosphere for 15 minutes at a temperature of 500.degree.C.
While the chemical oxide has been disclosed as an insulating layer
in an MNOS structure involving enhancement mode operation of a
P-channel device, other variations are possible. The chemical oxide
layer may also be used in depletion MNOS devices in a P-channel
configuration as well as in enhancement mode devices in an
N-channel tetrode arrangement (using Schottky injection). The
chemical oxide may also be used in MAOS devices wherein the A
stands for Al.sub.2 O.sub.3. One such device is an N-channel device
using enhancement mode operation. The chemical oxide may also be
used in a P-channel device operating in a depletion mode with
either a tunneling or avalanche injection mode of charging. In
general, the formation of aluminum oxide devices is well known and
follows the same general process as herein described in the
fabrication of an MNOS device. The nitride reaction in particular
is replaced by hydrolyzing AlCl.sub.3 in a mixture of H.sub.2 and
CO.sub.2 at 900.degree.C to form the Al.sub.2 O.sub.3 layer. The
dimensioning of the Al.sub.2 O.sub.3 layer follows the same
principles used in the Si.sub.3 N.sub.4 layer. The Al.sub.2 O.sub.3
layer thus is substantially thicker than the chemical oxide
>300A, and lies in the same range of thicknesses as the nitride
layer.
* * * * *